Electron Holography

Join the Competition Contribute to Benchmark

Standard reconstruction benchmark — forward model perfectly known, no calibration needed. Score = 0.5 × clip((PSNR−15)/30, 0, 1) + 0.5 × SSIM

#	Method	Score	PSNR (dB)	SSIM	Source
🥇	DiffHolo DiffHolo Gao et al. 2024 39.2 dB SSIM 0.953 Checkpoint unavailable	0.880	39.2	0.953	✓ Certified	Gao et al. 2024
🥈	PhysHolo PhysHolo Chen et al. 2024 37.8 dB SSIM 0.942 Checkpoint unavailable	0.851	37.8	0.942	✓ Certified	Chen et al. 2024
🥉	SwinHolo SwinHolo Wang et al. 2023 36.5 dB SSIM 0.931 Checkpoint unavailable	0.824	36.5	0.931	✓ Certified	Wang et al. 2023
4	TransHolo TransHolo Li et al. 2022 34.9 dB SSIM 0.913 Checkpoint unavailable	0.788	34.9	0.913	✓ Certified	Li et al. 2022
5	DeepHolo DeepHolo Rivenson et al. 2018 32.4 dB SSIM 0.875 Checkpoint unavailable	0.728	32.4	0.875	✓ Certified	Rivenson et al. 2018
6	DnCNN-Holo DnCNN-Holo Gao et al. 2019 29.6 dB SSIM 0.835 Checkpoint unavailable	0.661	29.6	0.835	✓ Certified	Gao et al. 2019
7	TV-Phase TV-Phase Beleggia et al. 2004 26.8 dB SSIM 0.783 Try in SpecLab →	0.588	26.8	0.783	✓ Certified	Beleggia et al. 2004
8	WDD-Holo WDD-Holo Lichte 1986 24.2 dB SSIM 0.742 Try in SpecLab →	0.524	24.2	0.742	✓ Certified	Lichte 1986
9	FFT-Holo FFT-Holo Lehmann & Lichte 2002 21.5 dB SSIM 0.700 Try in SpecLab →	0.458	21.5	0.700	✓ Certified	Lehmann & Lichte 2002

Dataset: PWM Benchmark (9 algorithms)

Blind Reconstruction Challenge — forward model has unknown mismatch, must calibrate from data. Score = 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖)

#	Method	Overall Score	Public PSNR / SSIM	Dev PSNR / SSIM	Hidden PSNR / SSIM	Trust	Source
🥇	SwinHolo + gradient SwinHolo + gradient Wang et al., Ultramicroscopy 2023 Score 0.751 Correct & Reconstruct →	0.751	0.826 35.37 dB / 0.970	0.742 30.87 dB / 0.929	0.685 26.88 dB / 0.855	✓ Certified	Wang et al., Ultramicroscopy 2023
🥈	DiffHolo + gradient DiffHolo + gradient Gao et al., NeurIPS 2024 Score 0.742 Correct & Reconstruct →	0.742	0.837 37.02 dB / 0.978	0.730 30.21 dB / 0.920	0.659 26.79 dB / 0.853	✓ Certified	Gao et al., NeurIPS 2024
🥉	PhysHolo + gradient PhysHolo + gradient Chen et al., Nat. Commun. 2024 Score 0.737 Correct & Reconstruct →	0.737	0.841 36.49 dB / 0.976	0.704 28.67 dB / 0.894	0.667 26.11 dB / 0.835	✓ Certified	Chen et al., Nat. Commun. 2024
4	TransHolo + gradient TransHolo + gradient Li et al., Nat. Commun. 2022 Score 0.721 Correct & Reconstruct →	0.721	0.805 33.44 dB / 0.956	0.718 29.09 dB / 0.902	0.640 25.5 dB / 0.818	✓ Certified	Li et al., Nat. Commun. 2022
5	DeepHolo + gradient DeepHolo + gradient Rivenson et al., Optica 2018 Score 0.678 Correct & Reconstruct →	0.678	0.768 30.99 dB / 0.931	0.661 25.56 dB / 0.819	0.606 23.46 dB / 0.749	✓ Certified	Rivenson et al., Optica 2018
6	DnCNN-Holo + gradient DnCNN-Holo + gradient Gao et al., Ultramicroscopy 2019 Score 0.634 Correct & Reconstruct →	0.634	0.727 28.57 dB / 0.892	0.603 23.58 dB / 0.753	0.571 22.5 dB / 0.711	✓ Certified	Gao et al., Ultramicroscopy 2019
7	WDD-Holo + gradient WDD-Holo + gradient Lichte, Ultramicroscopy 1986 Score 0.530 Correct & Reconstruct →	0.530	0.610 23.14 dB / 0.737	0.504 19.75 dB / 0.587	0.475 18.78 dB / 0.539	✓ Certified	Lichte, Ultramicroscopy 1986
8	TV-Phase + gradient TV-Phase + gradient Beleggia et al., Ultramicroscopy 2004 Score 0.530 Correct & Reconstruct →	0.530	0.668 25.5 dB / 0.818	0.501 19.97 dB / 0.597	0.420 17.73 dB / 0.487	✓ Certified	Beleggia et al., Ultramicroscopy 2004
9	FFT-Holo + gradient FFT-Holo + gradient Lehmann & Lichte, Microsc. Microanal. 2002 Score 0.459 Correct & Reconstruct →	0.459	0.534 20.28 dB / 0.612	0.440 17.54 dB / 0.477	0.403 16.31 dB / 0.416	✓ Certified	Lehmann & Lichte, Microsc. Microanal. 2002

Complete score requires all 3 tiers (Public + Dev + Hidden).

Join the competition →

Scoring: 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖) PSNR 40% · SSIM 40% · Consistency 20%

Public 3 scenes

Full-access development tier with all data visible.

What you get & how to use

What you get: Measurements (y), ideal forward operator (H), spec ranges, ground truth (x_true), and true mismatch spec.

How to use: Load HDF5 → compare reconstruction vs x_true → check consistency → iterate.

What to submit: Reconstructed signals (x_hat) and corrected spec as HDF5.

Public Leaderboard

#	Method	Score	PSNR	SSIM
1	PhysHolo + gradient	0.841	36.49	0.976
2	DiffHolo + gradient	0.837	37.02	0.978
3	SwinHolo + gradient	0.826	35.37	0.97
4	TransHolo + gradient	0.805	33.44	0.956
5	DeepHolo + gradient	0.768	30.99	0.931
6	DnCNN-Holo + gradient	0.727	28.57	0.892
7	TV-Phase + gradient	0.668	25.5	0.818
8	WDD-Holo + gradient	0.610	23.14	0.737
9	FFT-Holo + gradient	0.534	20.28	0.612

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
biprism_voltage	-2.0	4.0	V
fringe_spacing	-0.1	0.2	nm
partial_coherence	-5.0	10.0	%

View Details →

Dev 3 scenes

Blind evaluation tier — no ground truth available.

What you get & how to use

What you get: Measurements (y), ideal forward operator (H), and spec ranges only.

How to use: Apply your pipeline from the Public tier. Use consistency as self-check.

What to submit: Reconstructed signals and corrected spec. Scored server-side.

Dev Leaderboard

#	Method	Score	PSNR	SSIM
1	SwinHolo + gradient	0.742	30.87	0.929
2	DiffHolo + gradient	0.730	30.21	0.92
3	TransHolo + gradient	0.718	29.09	0.902
4	PhysHolo + gradient	0.704	28.67	0.894
5	DeepHolo + gradient	0.661	25.56	0.819
6	DnCNN-Holo + gradient	0.603	23.58	0.753
7	WDD-Holo + gradient	0.504	19.75	0.587
8	TV-Phase + gradient	0.501	19.97	0.597
9	FFT-Holo + gradient	0.440	17.54	0.477

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
biprism_voltage	-2.4	3.6	V
fringe_spacing	-0.12	0.18	nm
partial_coherence	-6.0	9.0	%

Submit Reconstruction View Details →

Hidden 3 scenes

Fully blind server-side evaluation — no data download.

What you get & how to use

What you get: No data downloadable. Algorithm runs server-side on hidden measurements.

How to use: Package algorithm as Docker container / Python script. Submit via link.

What to submit: Containerized algorithm accepting y + H, outputting x_hat + corrected spec.

Hidden Leaderboard

#	Method	Score	PSNR	SSIM
1	SwinHolo + gradient	0.685	26.88	0.855
2	PhysHolo + gradient	0.667	26.11	0.835
3	DiffHolo + gradient	0.659	26.79	0.853
4	TransHolo + gradient	0.640	25.5	0.818
5	DeepHolo + gradient	0.606	23.46	0.749
6	DnCNN-Holo + gradient	0.571	22.5	0.711
7	WDD-Holo + gradient	0.475	18.78	0.539
8	TV-Phase + gradient	0.420	17.73	0.487
9	FFT-Holo + gradient	0.403	16.31	0.416

Spec Ranges (3 parameters)

Parameter	Min	Max	Unit
biprism_voltage	-1.4	4.6	V
fringe_spacing	-0.07	0.23	nm
partial_coherence	-3.5	11.5	%

Submit Algorithm View Details →

Blind Reconstruction Challenge

Challenge

Given measurements with unknown mismatch and spec ranges (not exact params), reconstruct the original signal. A method must be evaluated on all three tiers for a complete score. Scored on a composite metric: 0.4 × PSNR_norm + 0.4 × SSIM + 0.2 × (1 − ‖y − Ĥx̂‖/‖y‖).

Input

Measurements y, ideal forward model H, spec ranges

Output

Reconstructed signal x̂

About the Imaging Modality

Off-axis electron holography records the interference pattern between an object wave (passed through the specimen) and a reference wave (passed through vacuum) using an electrostatic biprism. The hologram encodes the phase shift imparted by electric and magnetic fields within the specimen. Fourier filtering isolates the sideband carrying the complex wave information, from which amplitude and phase are extracted. Phase sensitivity of ~2*pi/1000 enables mapping of nanoscale electric and magnetic fields in materials.

Principle

Electron holography uses the interference between an object wave (transmitted through the specimen) and a reference wave (passing through vacuum) to record both amplitude and phase of the electron wave. An electrostatic biprism (charged wire) deflects the two waves to overlap and form interference fringes. Numerical reconstruction recovers the phase shift, which is sensitive to electrostatic potentials and magnetic fields in the specimen.

How to Build the System

Use a TEM (≥200 kV, FEG source for high coherence) equipped with an electron biprism (a thin metallized quartz fiber at adjustable voltage 50-300 V). Position the specimen so one half of the biprism overlaps the specimen edge and the other half is in vacuum. Record the hologram on a direct-electron detector. Fringe spacing should be 3-4× the desired resolution. Acquire reference holograms (empty) for normalization.

Common Reconstruction Algorithms

Fourier filtering (sideband extraction and inverse FFT for phase/amplitude)
Phase unwrapping for large phase shifts (>2π)
Mean inner potential measurement from phase maps
Magnetic induction mapping (from phase gradient of Lorentz holography)
In-line holography (through-focus series) with transport-of-intensity equation

Common Mistakes

Biprism voltage too low, giving insufficient overlap and poor fringe contrast
Fresnel fringes from specimen edge contaminating the holographic fringes
Not acquiring and dividing by a reference hologram, leaving biprism distortions
Specimen too thick, reducing fringe visibility from inelastic scattering
Stray magnetic fields causing unwanted phase shifts in the reference wave

How to Avoid Mistakes

Optimize biprism voltage for 3-4× oversampling of desired resolution with good contrast
Extend vacuum reference beyond the specimen edge; mask Fresnel fringe regions
Always acquire reference holograms and compute the normalized phase
Use thin specimens (< 50-80 nm) to maintain fringe contrast above 10%
Enclose the TEM column in mu-metal shielding; degauss the objective lens for Lorentz mode

Forward-Model Mismatch Cases

The widefield fallback produces real-valued output, but electron holography records the interference between object and reference electron waves — the complex-valued hologram encodes electromagnetic potentials (electric and magnetic fields) inside the specimen via the Aharonov-Bohm phase shift
The biprism interference fringes encode quantitative phase information (phase shift = C_E * integral(V(x,y,z)dz) for electrostatic, and -(e/hbar) * integral(A*dl) for magnetic) — the widefield blur destroys fringe contrast and all phase information

How to Correct the Mismatch

Use the electron holography operator that models biprism-mediated interference between object wave (with Aharonov-Bohm phase shift) and vacuum reference wave, producing complex holographic fringes
Reconstruct phase maps using Fourier sideband filtering and inverse FFT; for magnetic specimens, use Lorentz mode and separate electrostatic and magnetic phase contributions

Experimental Setup — Signal Chain

Experimental setup diagram for Electron Holography

Reconstruction Gallery — 4 Scenes × 3 Scenarios

Method: CPU_baseline | Mismatch: nominal (nominal=True, perturbed=False)

I Ideal II Mismatch (uncorrected) III Oracle correction

I — Ideal

Ground Truth

Measurement

Reconstruction

19.401488157362497 dB SSIM 0.48886704895877836

II — Mismatch

Ground Truth

Measurement

Reconstruction

16.857966791337656 dB SSIM 0.08915507096515726

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

19.015339082048712 dB SSIM 0.19323876033569679

I — Ideal

Ground Truth

Measurement

Reconstruction

20.386994971632934 dB SSIM 0.49771035050842166

II — Mismatch

Ground Truth

Measurement

Reconstruction

17.188182006293506 dB SSIM 0.08456095595654231

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

19.00326347865052 dB SSIM 0.1825080566719887

I — Ideal

Ground Truth

Measurement

Reconstruction

20.109288125342772 dB SSIM 0.4612187812713385

II — Mismatch

Ground Truth

Measurement

Reconstruction

17.552917796407378 dB SSIM 0.08129459333217216

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

19.721316784917846 dB SSIM 0.17921822573660165

I — Ideal

Ground Truth

Measurement

Reconstruction

19.33896725172696 dB SSIM 0.48831334454774855

II — Mismatch

Ground Truth

Measurement

Reconstruction

16.902556624008536 dB SSIM 0.09105537055754624

III — Oracle

Ground Truth

Measurement (perturbed)

Reconstruction

18.83727554088228 dB SSIM 0.1898195007747968

Mean PSNR Across All Scenes

19.80918462651629 dB

I — Ideal

17.125405804511768 dB

II — Mismatch

19.144298721624843 dB

III — Oracle

Degradation: -2.7 dB Recovery: +2.0 dB

Per-scene PSNR breakdown (4 scenes)

Scene	I (PSNR)	I (SSIM)	II (PSNR)	II (SSIM)	III (PSNR)	III (SSIM)
scene_00	19.401488157362497	0.48886704895877836	16.857966791337656	0.08915507096515726	19.015339082048712	0.19323876033569679
scene_01	20.386994971632934	0.49771035050842166	17.188182006293506	0.08456095595654231	19.00326347865052	0.1825080566719887
scene_02	20.109288125342772	0.4612187812713385	17.552917796407378	0.08129459333217216	19.721316784917846	0.17921822573660165
scene_03	19.33896725172696	0.48831334454774855	16.902556624008536	0.09105537055754624	18.83727554088228	0.1898195007747968
Mean	19.80918462651629	0.48402738132157175	17.125405804511768	0.0865164977028545	19.144298721624843	0.186196135879771

Experimental Setup

Instrument: Thermo Fisher Titan Holography / JEOL JEM-3000F

Accelerating Voltage Kv: 300

Wavelength Pm: 1.97

Detector: Gatan Orius CCD (2k x 2k)

Biprism Voltage V: 150

Exposure S: 2

Fringe Spacing Nm: 0.3

Reconstruction: Fourier filtering + phase unwrapping

Application: magnetic / electric field mapping

Key References

Dunin-Borkowski et al., 'Electron holography of nanostructured materials', Encyclopedia of Nanoscience and Nanotechnology (2004)
Lichte & Lehmann, 'Electron holography — basics and applications', Rep. Prog. Phys. 71, 016102 (2008)

Canonical Datasets

Holography benchmark datasets (Forschungszentrum Julich)

Spec DAG — Forward Model Pipeline

P(e⁻) → Σ(interference) → D(g, η₁)

P Electron Biprism (e⁻)

Σ Interference Sum (interference)

D CCD Camera (g, η₁)

Mismatch Parameters

Symbol	Parameter	Description	Perturbed
ΔV_b	biprism_voltage	Biprism voltage error (V)	2.0
Δd_f	fringe_spacing	Fringe spacing error (nm)	0.1
Δμ	partial_coherence	Spatial coherence error (%)	5.0

Credits System

40%

Platform Profit Pool

Revenue allocated to benchmark rewards

30%

Winner Share

Top algorithm receives from pool

$100

Min Withdrawal

Minimum payout threshold

Spec Primitives Reference (11 primitives)

P Propagation

Free-space or medium propagation kernel (Fresnel, Rayleigh-Sommerfeld).

M Mask / Modulation

Spatial or spatio-temporal amplitude modulation (coded aperture, SLM pattern).

Π Projection

Geometric projection operator (Radon transform, fan-beam, cone-beam).

F Fourier Sampling

Sampling in the Fourier / k-space domain (MRI, ptychography).

C Convolution

Shift-invariant convolution with a point-spread function (PSF).

Σ Summation / Integration

Summation along a physical dimension (spectral, temporal, angular).

D Detector

Sensor readout with gain g and noise model η (Gaussian, Poisson, mixed).

S Structured Illumination

Patterned illumination (block, Hadamard, random) applied to the scene.

W Wavelength Dispersion

Spectral dispersion element (prism, grating) with shift α and aperture a.

R Rotation / Motion

Sample or gantry rotation (CT, electron tomography).

Λ Wavelength Selection

Spectral filter or monochromator selecting a wavelength band.